VAOT: Vessel-Aware Optimal Transport for Retinal Fundus Enhancement

2025-11-23

AI Summary

In unpaired fundus image enhancement, generative adversarial network (GAN)-based methods often distort vascular structures, compromising topological connectivity and endpoint integrity. To address this, we propose the first optimal transport (OT)-based structural-preserving enhancement framework. Our method jointly incorporates a skeleton loss—ensuring global vascular connectivity—and an endpoint-aware loss—preserving local terminal stability—into the OT regularization objective, enabling anatomically consistent denoising and contrast enhancement. By integrating adversarial training, skeleton extraction, and endpoint detection, the approach operates without paired data. Evaluated on a synthetically degraded fundus dataset, it significantly outperforms state-of-the-art unpaired methods and improves downstream vascular and lesion segmentation performance by +3.2% in Dice score. The source code is publicly available.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Search and Optimization: Non-convex Optimization
• Machine Learning: Adversarial Learning & Robustness

Application Category

• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Security and Privacy: Privacy-enhancing technologies
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Color fundus photography (CFP) is central to diagnosing and monitoring retinal disease, yet its acquisition variability (e.g., illumination changes) often degrades image quality, which motivates robust enhancement methods. Unpaired enhancement pipelines are typically GAN-based, however, they can distort clinically critical vasculature, altering vessel topology and endpoint integrity. Motivated by these structural alterations, we propose Vessel-Aware Optimal Transport ( extbf{VAOT}), a framework that combines an optimal-transport objective with two structure-preserving regularizers: (i) a skeleton-based loss to maintain global vascular connectivity and (ii) an endpoint-aware loss to stabilize local termini. These constraints guide learning in the unpaired setting, reducing noise while preserving vessel structure. Experimental results on synthetic degradation benchmark and downstream evaluations in vessel and lesion segmentation demonstrate the superiority of the proposed methods against several state-of-the art baselines. The code is available at https://github.com/Retinal-Research/VAOT

Problem

• Enhancing retinal fundus images with variable acquisition quality
• Preserving vessel topology and endpoint integrity during enhancement
• Reducing noise while maintaining vascular structure in unpaired settings

Innovation

• Vessel-Aware Optimal Transport combines optimal-transport objective
• Skeleton-based loss preserves global vascular connectivity
• Endpoint-aware loss stabilizes local vessel termini